Since a few years, the Renault Group has been pursuing a digitalization strategy through its Industry 4.0 plan. Several digitalization initiatives have been implemented in some casting facilities, such as real-time data being transmited to the cloud from the HPDC press. However, most other data is neither centralized nor digitized, particularly data related to quality control. A project aimed at providing a framework for the digitalization of the casting operations with structured objectives has thus been created. This project has been organized into several main stages, the first of which involves creating the digital twin of the produced parts, based on tomographic scans. An X-ray tomograph is currently used to analyze the material integrity of these parts. Several studies have already focus on the extraction of features from a volume image of industrial parts, based on segmentation or AI [1, 2]. While the latter rely on massive database of defects and features which are either unavailable or hard to build, the first one suffers from image quality and artifacts which are hard to correct. In this approach, we chose to take advantage of our knowledge of the geometry and material composition of the produced part to create virtual radiographs which are comparable to the images obtained using a tomograph [3]. A geometric representation of the part is coupled with a physically-based model of the X-ray attenuation in order to simulate the radiographic images in a virtual tomograph. A 3D reconstruction of this geometric digital twin can thus be created using the same algorithm as the one used to for the real object which allows to generate 3D data containing the same artifacts in both volume representation. This set of 2 volumes can then be compared in terms of intensity. As only the real object contains porosities and geometric defects, they are naturally highlighted in the difference between the two volumes which can be processed, voxelized and meshed to obtain a suitable representation of the porosities and dimensional defects to improve our digital twin of the casted part.